96 Gabi Danon
be identified. One approach is to simply dismiss this as a marginal and exceptional
phenomenon found in only a few highly restricted domains or constructions. The
other is to assume that lexical semantics may ‘override’ the morphological features
of a noun as they enter into syntactic computation, thus giving rise to normal syn-
tactic agreement based on semantically determined features. These two approaches
make opposite predictions regarding the degree of productivity of this phenomenon.
According to the first approach, ‘semantic agreement’ should be extremely restricted
and construction-specific; according to the second approach, we might expect this
semantic ‘feature rewriting’ operation to be quite productive if it is an active part of
the grammar.
This paper argues that both of these approaches are mostly wrong, at least under
the kind of simple and naïve implementation often implicitly assumed. At the same
time, I argue that the two approaches actually do capture some aspect of the phenom-
enon correctly. One reason why two seemingly contradictory approaches might in fact
not exclude each other is that ‘semantic agreement’ is not a uniform and homogeneous
set of facts. In this paper I focus on two distinct cases: plural subjects that seem to trig-
ger singular agreement, and singular subjects triggering plural agreement. I will argue
that underlying these two cases, both of which have sometimes been seen as semantic
agreement, are two mechanisms that differ from each other in some crucial ways.
A major aspect of understanding these phenomena is to identify the exact proper-
ties of the constructions at hand. A significant part of this paper is devoted to showing
that descriptively, these two types of agreement mismatch have different distribu-
tional properties and are subject to different constraints. This forms the basis for the
two analyses to be proposed, which are meant to capture in an explicit manner the
observed facts and to show that neither of these two phenomena poses a real prob-
lem to modularity and to the hypothesis that agreement is a syntactic operation that
semantics does not directly ‘interfere’ with. In particular, it is argued, following Danon
(2012), that singular ‘agreement’ with plural subjects is not agreement at all (whether
semantic or syntactic) but lack of agreement, whereas plural agreement with singular
subjects is the result of regular syntactic agreement.
I will propose that lack of agreement entails that no thematic role can be assigned
to the subject, and hence the distribution of non-agreeing subjects is limited to envi-
ronments where a non-thematic subject is allowed. Plural agreement with singulars,
on the other hand, will be argued to be the result of a lexical mismatch between two
bundles of features, where the relevant nouns carry both singular and plural features at
the same time. The analysis of both types of agreement mismatch depends crucially on
adopting a model of nominal features that would allow us to distinguish between the
features used for noun phrase internal concord and those used for external agreement;
building upon previous work in Danon (2012, 2013 ), I will argue for a dual analysis
based on the insights of Wechsler and Zlatić (2000, 2003 ).